08/18/2026
For years, enterprise data teams were solving one central problem:
Can humans find and trust the right data?
That question shaped the first era of data catalogs, glossaries, lineage, ownership, and quality workflows.
But now, the buyer's question is changing:
Can AI agents find, trust, and safely use the right data?
AI agents are changing the interface between enterprise users and enterprise systems.
Earlier, a human searched the catalog, opened the dashboard, checked the glossary, interpreted the metric, and decided what to trust.
Now, an agent may be the one discovering tools, calling systems, retrieving context, and triggering downstream work.
This is where Anthropic’s Model Context Protocol becomes important.
MCP gives AI systems a standard way to connect with external tools, data sources, and business systems.
But the protocol is only one part of the story.
The more important enterprise idea is the MCP server.
An MCP server is the layer that exposes specific tools, data, and capabilities to the agent in a structured way.
It tells the agent:
What tools are available.
What each tool can do.
How the tool should be called.
What information can be returned.
So the MCP server becomes a new interface between the agent and the enterprise.
And that creates the next set of challenges.
It is not enough for the MCP server to expose “a revenue table.”
It needs to expose the trusted revenue definition.
It is not enough to expose “customer data.”
It needs to expose the permitted customer data for that agent, user, and task.
It is not enough to return “an answer.”
It needs to return an answer with trust, lineage, freshness, and policy context.
This is where the role of governance changes.
In the human era, governance helped people find, understand, and trust data.
In the agent era, governance has to shape what agents can see, understand, use, and act on.
That is a very different job.
Humans often compensate for weak governance.
A business analyst may notice when a number feels off.
A data steward may know which source is trusted.
A finance leader may question whether a metric matches the official KPI.
Agents do not automatically carry that business memory.
If an MCP server exposes a weak, stale, or poorly governed context, the agent may still use it confidently and create downstream risk.
The old problem was human discovery.
The new problem is agent-safe consumption.
MCP servers may become the interface.
But the governed context will decide whether agents can safely act.
